Basic Image Processing (using ImageJ)

Basic Image Processing (using ImageJ) Dr. Arne Seitz Swiss Institute of Technology (EPFL) Faculty of Life Sciences Head of BIOIMAGING AND OPTICS BIOP arne.seitz@epfl.ch

Overview File formats (data storage) Programs for image viewing / processing / representation Basic Image Processing (using ImageJ)

Definition Digital image A digital image is a representation of a twodimensional image using ones and zeros (binary). (Wikipedia) Analog = continuous values Digital = discrete steps 1 1 1 1 1 0 0 0 1 1 1 0 1 0 0 0 1 1 01 1

Detection Devices Array detector Point detector

File Formats data storage Lossless image formats Lossy compression formats Custom formats (microscope companies) Sequence vs. single image per file 8bit, 12bit, 16bit, 32bit, RGB Storage: Always have at least 1 copy of the data Very suitable fileservers (automatic backup)

Lossless Image Formats TIFF (with our without compression) BMP (windows uncompressed) GIF (graphics interchange format) PNG (portable network graphics) Raw data text image Microscopy Primer http://micro.magnet.fsu.edu/primer

Image Format: TIFF Tag Image File Format Image header with flexible set of tags which can be used to store e.g. microscopic settings Flexible in color space and bit depth Microscopy: grayscale 8bit, 16 bit (12bit data) Color (e.g. Overlay): RGB (red green blue 8bit each) Quantification: 32bit (floating point values) Always lossless: Uncompressed or compressed Multiple images possible in one file

Image Compression: TIFF Run Length Coding (RLE): first number discribes the color, the second the number of following pixels having the same color. (0,0,0), 6 (0,255,0), 9 (0,0,0), 2 (255,0,0), 6 LZW (Lempel-Ziv-Welch): Find repetative patterns of values and give them a number which is points to an entry of a dictionary (LUT). 1 2 3

Image Compression: TIFF Pros: Extra infos can be written in the tags (e.g. microscope data like objective lens, voxel size) Everybody can read it Lossless Flexible (8, 16, 32bit grayscale, 8:8:8bit RGB) 256 Floating point values 65536 graylevels Cons: Big files Compressed files can t be loaded by ImageJ

Lossy Image Formats The lossy compression algorithm takes advantage of the limitations of the human visual senses and discards information that would not be sensed by the eye. (like mp3 in audio). Compression level is usually flexible, but the more compressed the more information is lost and artifacts become visible by eye From: www.wikipedia.org

Image Compression: JPG Split image into color and gray-scale information (color is less important than bounderies) reduce high frequency color information. Group pixel into 8x8 blocks and transform through discrete cosine transform

Image Compression: JPG Pros: Small Files True Color Usable for most photos (real life) and presentations (powerpoint) Cons: Do not use for quantification! Unrelevant photoinfos get lost Every file-saving reduces the quality

Image Viewers ImageJ (Java based, freeware, Win/MAC/Linux) Irfanview (www.irfanview.com/) Freeware Convert (e.g. tif jpg) Batch processing ACDSee (ACD Systems) Microscope companies Zeiss Image Browser / Axiovision LE Leica LCS Lite Olympus Viewer

Image Representation ImageJ Imaris (Bitplane): 4 floating licenses installed on image processing workstations Photoshop, Paintshop, Illustrator, Corel Draw (, Powerpoint) Volocity (Improvision): Custom software of microscopes

Image Processing ImageJ (http://rsb.info.nih.gov/ij/index.html) installed on all image processing workstations Installation: http://pacific.mpi-cbg.de/wiki/index.php (Fiji=ImageJ+plugins+regular update) Manual: www.uhnresearch.ca/facilities/wcif/imagej/ (also available as pdf) Additional plugins: http://rsb.info.nih.gov/ij/plugins/index.html Metamorph (Universal Imaging), installed on 2 image processing workstations Custom software of microscopes

Image Processing Basics Visual Image Inspection Lookup tables (LUT) and LUT operations Histogram, brightness, contrast Filter Threshold Measurements Color functions

Visual Image Inspection Displaying images, histogram Microscopy Primer http://micro.magnet.fsu.edu/primer

Visual Image Inspection Displaying images, histogram Pixel count Intensity value

Lookup table (LUT) LUT operations Displays can only show 256 gray values (8bit) per color Data is unchanged, it s only mapped differently Data Intensity Displayed Intensity 0 0 179 0 180 5 181 10 226 227 228 255 229 255 65535 255

Brightness, Contrast Caution: Apply modifies the data! Dr. Arne Seitz Contrast is the difference in visual properties that makes an object distinguishable from other objects and the background.

Color LUT The pixel contains a pointer to an array, where the actual pixel values are stored old LUT: 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1 1 3 3 3 3 3 3 1: (0,102,255) 2: (51,102,255 3: (10,100,200) 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1 1 3 3 3 3 3 3 new LUT: 1: (0,0,0) 2: (0,255,0) 3: (255,0,0) Dr. Arne Seitz HiLo LUT

Color LUT The pixel contains a pointer to an array, where the actual pixel values are stored old LUT: 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1 1 3 3 3 3 3 3 1: (0,102,255) 2: (51,102,255 3: (10,100,200) 1 1 1 1 1 1 2 2 2 2 2 2 2 2 2 1 1 3 3 3 3 3 3 new LUT: 1: (0,0,0) 2: (0,255,0) 3: (255,0,0) Dr. Arne Seitz Rainbow LUT

Non-linear Histogram Stretch Enhance contrast by (changing data): Equalization non-linear stretch based on square root of the intensity Raw data Linear stretch Normalization

Equalization

Gamma Gamma is a non-linear histogram adjustment 8 bit images: New intensity = 255 [(old intensity/255) gamma]

Filtering Image processing filters are mainly used to: suppress the high frequencies in the image, i.e. smoothing the image, noise reduction or suppress the low frequencies, i.e. enhancing or detecting edges in the image An image can be filtered either in the frequency or in the spatial domain. Filtering in the frequency domain requires Fourier transform first and re-transformation after application of the filter. Filtering in the spatial domain is done by convolving the image with the filterfunction.

Filtering Shifting and multiplying a filter kernel Filtered image

Noise Reduction: Mean 1 9 1 9 1 9 1 9 1 9 1 9 mean 1 9 1 9 1 9 Mean 1pt

Noise Reduction: Gaussian Filtering with a gaussian bell-shaped kernel: 1 2 1 10 25 3 9 33 5 1 16 2 4 2 1 2 1 1 16 10 50 3 18 132 10 4 6 8 4 12 8

Noise Reduction: Median Median 3x3 median Median 5x5

Noise Reduction: Median, Mean Median, 1pt Mean, 1pt

Median-, Mean-, Max-, Min-Filter Median, 5pt Min, 2pt Mean, 5pt Max, 2pt

Mean-, Gauss-Filter Mean, 2pt, 4 pt Gauss, 2pt, 4 pt

Mean-, Median-Filter Mean, 2pt, 4 pt Median, 2pt, 4 pt

Min-, Max-Filter Min, 2pt Max, 2pt

-1-1 -1-1 9-1 -1-1 -1 Sharpen / Blur 1 1 1 1 2 1 1 1 1 sharpen blurring

Example: Edge-Finding with derivatives -1-1 -1 0 0 0 1 1 1-1 -1-1 0 1 0 1 1 1

Even background: Background Subtraction subtract average background from image Subtract background image (same exposure time without illumination) Uneven background: Rolling ball filter Use kernel larger than diameter of largest object Original Image Opening Original Image - Opening

Line Profile Without background subtraction After rolling ball (50) background subtraction

Thresholding Thresholding is used to change pixel values above or below a certain intensity value (threshold): Threshholding is a simple method for Segmentation (separation and location of objects of interest)

Measuring Sizes Set Scale with pixel (voxel) size Include Scalebar

Measuring Length

Area Measurement 16bit image 32bit image 32bit image, background thresholded to Not a Number 16bit image, same threshold as in 32bit image but not applied

Analyze Particles Segmented objects

Threshold and Opening/Closing dilate erode Dr. Arne Seitz Closing: Dilate/Erode Opening: Erode/Dilate

Color Functions RGB Merge /RGB Split

Deconvolution From Object to Image Effects causing Image degradation: Object Image Noise Signal derived noise Noise emerging from the digital imaging system Scatter Caused by heterogeneous refractive index (RI) Glare Random disturbance of light in the system Blur

Point Spread Function (PSF) A Point Spread Function is the 3D diffraction pattern of a point source of light. Widefield = hourglass shape Confocal = American Football shape

Convolution of an Object Object can be referred as accumulation of points Each point is visible as a PSF Object PSF = Image = convolution Image process hast to be - Linear - Shift invariant Convolution is in principle a reversible mathematical equation

Constrained Iterative Constrained: Nonnegativity Smoothing or regularization to suppress noise amplification Iterative: Best estimate is found in a successional serial of calculations.

lead to different Results raw data Different Algorithms AutoQuant: non blind 15 It AutoQuant: Blind 15 It Huygens: CMLE 30 It SoftWorx: 30 It

Signal improvement 7000 6000 5000 4000 1400 1200 1000 800 Higher signal to background ratio More distinct peaks 3000 600 2000 400 1000 200 0 AQ_blind_15It_thPSF not deconvolved 0

WF Deconvolution Computational substraction of blur or reassignment to the assumed source Advantages: Good light efficiency (esp. with reassignment) CCD instead of PMT (high Quantum efficiency) Fast stack recording possible low bleaching Disadvantages: Need for high computational systems Artefacts can not be excluded

WF Decon vs. Confocal To deconvolve or not to deconvolve That is not the question: WF + Deconvolution is no real alternative to Confocal pictures as they can also be deconvolved

Conclusions Keep environment constant and convenient Use powerful dyes Think about required resolution (x, y, z, t, brightness, channel number) to minimize photostress Use appropriate microscopy method

Summary Use lossless file formats for archiving important data Image processing is an important step in generating (optimal) results Only use documented image processing steps/routines

More about image processing 1. Lecture M. Unser, EPFL see also website: http://bigwww.epfl.ch/ 2. Books a) W. Burger, M. J. Burge Digital Image Processing, Springer 2008 b) J. C. Russ The image processing Handbook, CRC Press 2007 3. PT-BIOP EPFL, SV-AI 0241, SV-AI 0140 http://biop.epfl.ch/